View source for Reaktor race-condition

== Introduction ==

https://race-condition.reaktor.com/play

Input data 
* FPV view
* speed?
* best lap time, last lap time?

Output
* throttle
* steering

== Theory ==

=== Rules ===

=== PID algorithm ===

PID to stay in the blue area.

=== NEAT algorithm ===

Input
* Distance to some black; about 15. 

A.I. teaches itself to drive in Trackmania: https://www.youtube.com/watch?v=a8Bo2DHrrow
* The neat algorithm: https://neat-python.readthedocs.io/en/latest/neat_overview.html
* Input: Wall distances and speed (shown at t=1.01; 7 or 15 distances)
* 8 inputs (distances and speed) -> 3 outputs (left, right, throttle)
* Fitness function: distance traveled by the car in a given time
* Generation 1: Population of 100. 
* Generation n: crossovers of best individuals + random mutations
* Q-learning,

https://www.youtube.com/results?search_query=neat+algorithm+game

https://medium.com/@nirajsawant2313/a-simple-ai-using-neat-neuroevolution-of-augmenting-topologies-ae1b7d8f4a5e
* Inputs: The relative position of the ball to the paddle, velocity of the ball. Totally 4.
* Outputs: Left, Right, place
* Fitness function: +10 if hits, -5 if misses.


https://towardsdatascience.com/neuro-evolution-on-steroids-82bd14ddc2f6

=== Soft Actor-Critic Algorithm ===  

https://github.com/trackmania-rl/tmrl
* Soft Actor-Critic: a beginner-friendly introduction https://www.youtube.com/watch?v=LN29DDlHp1U